#asahi-dev · 2022-02-15 20:25
axboeissue on N, irq right after on N+1 - seems far fetched though#asahi-dev · 2022-02-15 20:25
axboesven: my worry was between CPUs#asahi-dev · 2022-02-15 20:23
axboesven: cuddle up with a cup of your favorite beverage and open Documentation/memory-barrier.txt ;)#asahi-dev · 2022-02-15 20:22
axboeI'd sugggest just going with the lock + irq disable around the issue for now#asahi-dev · 2022-02-15 20:22
axboethen again, not an ordering expert, I can always convince myself either way...#asahi-dev · 2022-02-15 20:22
axboebut not sure how that's even easily doable without serialization anyway#asahi-dev · 2022-02-15 20:22
axboewe _might_ just need an appropriate barrier so that it's visible on a different CPU#asahi-dev · 2022-02-15 20:20
axboeok#asahi-dev · 2022-02-15 20:20
axboehow relaxed is the arm memory model?#asahi-dev · 2022-02-15 20:20
axboethough not quite sure why it needs to be serialized against irq#asahi-dev · 2022-02-15 20:19
axboebut having unserialized issue seems like asking for trouble#asahi-dev · 2022-02-15 20:19
axboesven: trying to come up with a theory :)#asahi-dev · 2022-02-15 20:04
axboecould just be timing and I'll still hit it, but seemed to be trivial to hit before and seems solid now#asahi-dev · 2022-02-15 20:03
axboehttps://pastebin.com/PYTqnDrG#asahi-dev · 2022-02-15 20:02
axboegrabbing anv->lock around the issue does seem to fix it for me#asahi-dev · 2022-02-15 19:57
axboesven: yeah I guess as long as the writel is atomic it should be fine#asahi-dev · 2022-02-15 19:55
axboesven: question, what serializes writes to q->sq_db?#asahi-dev · 2022-02-15 19:55
axboesven: never saw it before adding that#asahi-dev · 2022-02-15 19:42
axboeeg loading firefox with bunch of tabs#asahi-dev · 2022-02-15 19:42
axboedoes look like it's correlated with some IO load and CPU usage#asahi-dev · 2022-02-15 19:40
axboe(after reset)#asahi-dev · 2022-02-15 19:40
axboeand it did come up fine#asahi-dev · 2022-02-15 19:40
axboeok so tag 2 this time, it reset the controller#asahi-dev · 2022-02-15 19:39
axboesame thing again#asahi-dev · 2022-02-15 19:39
axboeok it did#asahi-dev · 2022-02-15 19:39
axboedidn't reproduce now on starting tbird + ff#asahi-dev · 2022-02-15 19:38
axboeloading up the device with just reads doesn't lock it up#asahi-dev · 2022-02-15 19:37
axboehttps://kernel.dk/nvme.jpg#asahi-dev · 2022-02-15 19:34
axboelet me know if you want to see the pic#asahi-dev · 2022-02-15 19:34
axboecompletioned polled for all 3, on last one it resets the controller, reset fails, IO errors ensue#asahi-dev · 2022-02-15 19:34
axboehang on, took a pic#asahi-dev · 2022-02-15 19:34
axboeit does#asahi-dev · 2022-02-15 19:31
axboeboots fine, log into X and start a bunch of things#asahi-dev · 2022-02-15 19:31
axboe30 seconds between each#asahi-dev · 2022-02-15 19:31
axboetimes out tag 10, then 2, then 3 and resets controller#asahi-dev · 2022-02-15 19:30
axboehmm weird#asahi-dev · 2022-02-15 19:29
axboeoh sorry, io queue#asahi-dev · 2022-02-15 19:29
axboehaven't seen that before#asahi-dev · 2022-02-15 19:29
axboesven: two subsequent boots I've run into aq timeouts#asahi-dev · 2022-02-15 18:58
axboej`ey: if we ever end up at the same conference, I'll tell you the story of the time hch took my kids to watch the lego movie#asahi-dev · 2022-02-15 18:56
axboesven: I think the platform one is a good start#asahi-dev · 2022-02-15 18:54
axboeas long as they stay out of the fast path, then I don't really care#asahi-dev · 2022-02-15 18:54
axboeit's more of an hch'ism#asahi-dev · 2022-02-15 18:53
axboe(not a huge issue imho, it's not very invasive)#asahi-dev · 2022-02-15 18:52
axboeok, so we can't drop the generic quirks#asahi-dev · 2022-02-15 18:52
axboeah ok, dependencies#asahi-dev · 2022-02-15 18:51
axboesven: upstream would look kindly at the apple-nvme driver, as it can be combined with a "remove various weird apple workarounds or quirks" in the generic one#asahi-dev · 2022-02-15 18:51
axboesven: speaking of patches that could be pushed out... ;)#asahi-dev · 2022-02-15 18:50
axboeI've only once seen higher AQ depth be useful, and that was under provisioning and with discard sizes being smaller where then having a higher admin queue depth made a difference#asahi-dev · 2022-02-15 18:49
axboesven: as a temporary improvement, adjusting AQ to 2 would make sense imho